How should they be evaluated and what evidence is needed for WHO endorsement?

Transcription

1 New predictive tests for the diagnosis of tuberculosis infection How should they be evaluated and what evidence is needed for WHO endorsement? Frank Cobelens KNCV Tuberculosis Foundation The Hague, Netherlands Amsterdam Institute for Global Health and Development Amsterdam, Netherlands Cape Town, 3 Dec 2015

2 What is our target condition? PICO Among persons at high risk of LTBI who are not on tuberculosis preventive therapy, which test(s) alone or in combination with other proxies for LTBI, when positive, can best identify individuals most at risk of progression? New test has to be better predictor of TB than existing tests Questions: Why is better prediction important? what does better prediction mean?

3 NNT Number needed to treat 455 NNT to prevent 1 true case of TB using IGRA IGRA sensitivity 78% IGRA specificity 58% % 5% 10% 15% 20% 25% 30% Cumulative TB Prevalence incidence Kik & Cobelens, in prep

4 NNT Number needed to treat 455 NNT to prevent 1 true case of TB using IGRA IGRA sensitivity 78% IGRA specificity 58% % 5% 10% 15% 20% 25% 30% Cumulative TB Prevalence incidence HIV-positive contacts Kik & Cobelens, in prep

5 NNT Number needed to treat 455 NNT to prevent 1 true case of TB using IGRA IGRA sensitivity 78% IGRA specificity 58% % 5% 10% 15% 20% 25% 30% Cumulative TB Prevalence incidence HIV-negative contacts prisoners Kik & Cobelens, in prep

6 What does better prediction mean? Esmail et al. Phil Trans Roy Soc 2015

7 Can we at all predict disease? We do not know what precipitates TB disease in a latently infected individual therefore beyond existing risk classification we cannot predict who will and who will not become diseased (i.e. it s a stochastic event)

8 What does the test measure? Conceptually, the test either predicts that disease cannot happen because there is no persistent infection persistent infection test... or predicts that disease occurs because it has already started. incipient TB test

9 % free of incident TB 100 everyone with positive TST or IGRA at time of infection Time (months) since exposure

10 % free of incident TB Test for persistent infection TST cannot distinguish Non-persistent (cleared) infections Time (months) since exposure

11 % free of incident TB Test for persistent infection TST cannot distinguish Non-persistent (cleared) infections Time (months) since exposure

12 % free of incident TB Test for persistent infection TST cannot distinguish Non-persistent (cleared) infections Time (months) since exposure

13 % free of incident TB Test for persistent infection ????? Low progression rate High progression rate Non-persistent (cleared) infections Time (months) since exposure

14 % free of incident TB Test for persistent infection ???? PPV - NNT population dependent This is probably what current IGRA s do? Re-exposure rare Non-persistent (cleared) infections Re-exposure frequent Time (months) since exposure

15 So we need a test that predicts that... disease occurs because it has already started. incipient TB test

16 % free of incident TB Test for incipient TB ???? Low probability of being positive (high NNS) But if positive, high probability of disease (low NNT), largely independent of population tested Only expect disease over short time period Time (months) since exposure

17 % of incident TB cases Incipient TB: required length of follow-up 176 Chinese patients with abnormal X-rays but 5 negative cultures Followed up for TB for 36 months: 93 TB cases (69 culture-confirmed) all TB culture+ TB months 4-8 months 7-12 months months months months Time since initial X-ray Hongkong Chest Service. Am Rev Resp Dis 1981

18 A few consequences (1) For targeting preventive treatment we are not interested in latent TB infection as such, but in predicting disease WHO endorsement must be ultimately based on prediction of disease Some designs used in evaluation of IGRA will be non-informative: studies comparing test results with that of IGRA or TST as reference standard (beyond very early stages of test evaluation candidate selection) studies that analyze test results along a M.tb exposure gradient

19 A few consequences (2) The new test may identify the same absolute number of persons who develop TB disease as TST or IGRA but with much higher PPV (= lower number-needed-to-treat) Comparative studies cannot just have effectiveness endpoints but must also have cost-benefit endpoints Should this be required for WHO endorsement?

20 Evaluation phases 1. Analytical evaluation: evaluation of different subsets of well characterized (banked) samples 2. Clinical evaluation: evaluate the test in the intended target population in a controlled setting with high quality standards (compare the results of the new test against a reference standard) 3. Evaluation for (public) health impact: evaluate the test under routine conditions for impact on patient-important or health system-important outcomes (comparison against a reference standard not necessary) Also important: reproducibility, conversion/reversions, field robustness, feasibility, accepatbility etc. But I will not address them here

21 1. Analytical evaluation Key questions: 1. Is the test positive in persons with active TB? (at beginning of therapy) 2. Is the test negative in persons never exposed to M.tb? 3. Is the test positive in persons who develop active TB over months? 4. Is the test negative in persons who remain without active TB over same period? Design: Analysis of well-characterized banked specimens Limitations: Various, but this phase is primarily meant to select promising candidate assays for further development and evaluation

22 2. Clinical evaluation Key questions: 1. Is the test positive in persons who develop active TB over months? 2. Is the test negative in persons who remain without active TB oversame period? Design: Follow-up studies of persons at high risk of developing TB Various sub-populations, including people living with HIV, people with diabetes, malnutrition, children, elderly Options: 1. Cohort designs 2. Nested case-control designs 3. Case-base (case-cohort) designs

23 Cohort designs Test negative positive # TB cases # TB cases Follow tested individuals actively over up to 24 months Active ascertainment of incident TB, stratified by test result Requirements: Minimal cohort attrition Will be related to length of follow-up Blinded TB case ascertainment with regard to test result Re-infection rate should not be very high Re-infection reduces relative risk if occurred randomly

24 Cohort designs - issues Large sample size needed depends on expected TB incidence and length of follow-up identify high risk study populations Follow-up for relapse after TB treatment as a cohort design model TB ascertainment should be culture-confirmed leads to unbiased relative risk, but increases sample size

25 Nested case-control design Test negative positive TB cases + Non-TB cases Follow tested individuals passively over defined period (passive cohort) Passive ascertainment of incident TB Test status among incident TB cases compared to that of random subset of non-tb cases Allows for larger sample sizes Requirements: Probability of being included as a TB case should be independent of test result Blinded TB case ascertainment with regard to test result Re-infection rate should not be very high

26 Case-control designs - issues Some TB cases will be missed the cases in the analysis are assumed to be a random subset of all cases accrued in the cohort --> potential for selection bias Record linkage may lead to misclassification TB ascertainment should be culture-confirmed may be more difficult with passive follow-up Do not test all at t=0 but store specimens only

27 2. Evaluation of health impact Key questions: 1. Does the test when used in routine settings improve health outcomes? 2. Does the test when used in routine settings improve cost-effectiveness? Design: Comparative designs, ideally randomized trial (individual/group)

28 Comparative designs (trials) Old test Treat positives # TB cases Randomize New test Treat positives # TB cases In each arm follow individuals actively over time (treated + non-treated) Active ascertainment of incident TB, stratified by arm Determine: Incidence ratio and difference for TB Incidence ratio and difference for adverse events from preventive treatment Difference in number-needed-to-treat Cost-effectiveness

29 Conclusions What we re looking for is a test for incipient TB This requires a different evaluation approach than used for IGRA thus far Endorsement should ultimately be based on predictive power (of incident TB) follow-up studies Cohort studies with relatively short follow-up are needed for clinical evaluation Nested case-control studies may be useful alternative Randomized trials are ideally done to show impact on patient/health system-important outcomes For such trails, number-needed-to-treat, adverse events and cost-effectiveness are important endpoints

30 Thank you